Transcript 강의자료_손희정_20130325_1
Cross Sectional Studies Son Hee Jung 2013/03/25
Type of Epidemiological Studies Type of study Alternative name Unit
Experimental
RCT
Observational
Ecological Cross sectional Case-control Cohort clinical trial correlational prevalence case-reference follow up individuals population individuals individuals individuals
Study Designs & Corresponding Questions • Cross-sectional • Ecologic • Case-control • Prospective
How common is this disease or condition?
What explains differences between groups?
What factors are associated with having a disease? How many people will get the disease? What factors predict development?
Contents • • • • • • Definition Basic approach Advantage & disadvantage Sampling Measures of disease – Prevalence Bias
Cross-sectional study-definition
Cross Sectional Study 연구대상 집단 한 시점 연구 진행 요인 노출과 질환에 관한 정보 수집
Cross-sectional study- Characteristics
Basic approach • • • Include a sample of all persons in a population at a given time without regard to exposure or disease status Typically exposure and diseases assessed at that one time Exposure subpopulations can be compared with respect to disease prevalence
Basic approach • • • For some questions, temporal ordering between exposure and disease is clear and cross sectional studies can test hypothesis – Example: genotype, blood type When temporal ordering is not clear can be used to examine relations between exposure and outcomes descriptively, and to generate hypotheses Can combine a cross sectional study with follow up to create a cohort study
Basic approach • Issues with addressing etiology – Temporal ordering between exposure and outcome cannot be assured – Length biased sampling • Cases with long duration will be over represented
Cross -Sectional Studies: Advantages • • • • Inexpensive for common diseases Should be able to get a better response rate than other study designs Relatively short study duration Can be addressed to specific populations of interest
Cross-Sectional Studies : Disadvantages • • • • Unsuitable for rare or short duration diseases High refusal rate may make accurate prevalence estimates impossible More expensive and time consuming than case control studies No data on temporal relationship between risk factors and disease development
Why sample?
Sampling from the source population
• Non-probability sampling Common convenience sampling methods – Street surveys • Use convenient place such as mall, hospital – – Mail-out questionnaires • Most dangerous • Feel very strongly about the issue->bias Volunteer call • Selection bias
Non-probability sampling-Convenience sampling • • Select a sample through an easy, simple or inexpensive method Problem – – High risk of creating a bias May provide misleading information – Can be accepted, but… • • Be careful in assessing And the results they produce
Basic probability sampling • Simple random sampling – Each sample of the chosen size has the same probability of being selected
Basic probability sampling • Systematic sampling – Obtain a lost of an available population, ordered according to an unrelated factor – – Pick a number n as step size Pick every n-th subject of the list
Stratified random sampling
Cluster random sampling
Multistage sampling
The National Health and Nutrition Examination Survey (NHANES)
NHANES Interviews & Examinations • ㅍ
NHANES Sample Design
Analyses of NHANES Data
Weighting in NHANES • ㅍ
NHANES base probability of selection • ㅍ
Oversampling
Sample Weights
Why weight?
Probability weight – simple example
Example of weighting • • • • Imagine 100 male & 100 female in sample But only 80 males & 75 females respond Male respondent will get weight of – 100/80->1/(80/100)=1.25
Female respondent will get weight of – 100/75->1/(75/100)=1.33
국민건강영양조사의 표본추출방법 예
다단계 표본추출 • 단순무작위 표본추출의 실제적 어려움을 해결하 기 위해 고안된 방법 – 전국 규모의 여론조사에 이용 – “series” of simple random samples in stages random sampling • 국민건강영양조사 random sampling random sampling
유병률 산출 : 가중치 적용 • 목적 : 국민건강영양조사의 표본이 우리나라 국민 을 대표하도록 가중치를 사용
Direct age adjustment-before population 900,000 A No. of death 862 Death rates per 100,000 96 population 900,000 B No. of death 1,130 Death rates per 100,000 126 Age group All ages 30-49 50-69 70+ A population No. of death 900,000 862 Death rates per 100,000 96 500,000 300,000 60 396 12 132 100,000 406 406 B population No. of death 900,000 300,000 400,000 200,000 1,130 30 400 700 Death rates per 100,000 126 10 100 350
Direct age adjustment-after population 900,000 A No. of death 862 Death rates per 100,000 96 population 900,000 B No. of death 1,130 Death rates per 100,000 126 Age group Standard population All ages 30-49 50-69 70+ Total 1,800,000 800,000 700,000 300,000 Age-adjusted rates “A" age-specific mortality rates per 100,000 12 132 406 Expected No. of deaths using “A" rates “B" age-specific mortality rates per 10 0,000 Expected No. of d eaths using “B" rates 96 924 1,218 2,238 124.3
10 100 350 80 700 1,050 1,830 101.7
Age-adjusted rates: 2238/1800000=124.3 1830/1800000=101.7
• • • Indirect age adjustment (Standardized Mortality Ratio) When – – number of deaths for each age-specific strata are not available Study mortality in an occupational exposure population Defined Observed number of deaths per year SMR= Expected number of deaths per year X100 SMR of 100 • Observed number of deaths is the same as expected number of deaths
Sampling, Inference, and generalization
Sampling, Inference, and generalization
Sampling, Inference, and generalization
If you tell the truth you don't have to remember anything. by Mark Twain 1894
Why do we measure disease prevalence?
Measuring burden: prevalence
Prevalence
Measuring burden: prevalence
Person-time at risk: exposed and unexposed
Censored individuals
Censoring
Measuring of prevalence
Point and period prevalence: example
Point prevalence at several time points
Period prevalence
Lifetime prevalence Life time prevalence 4/5
Prevalence of diabetes
Utility of prevalence
Sloppy use of risk
Sloppy use of rate
Classic example of rate that is not a rate
Case fatality(rate?)
Proportional mortality (rate?)
Total deaths united states 2004
Deaths , U.S. 2004 ages 20-24 Years
What ‘s a possible inferential problem with proportional mortality?
Measuring risk: cumulative incidence
Measuring risk: cumulative incidence
Cumulative incidence is a proportion
Calculating the cumulative incidence
Odds
Odds
Odds
Odds
Odds and probabilities
• The higher the incidence, the higher the discrepancy.
Prevalence, Incidence, disease duration
Disease prevalence depends on
Incidence rates can be calculated for each transition in health status
Incidence rates can be calculated for each transition in health status
Relationship among prevalence, incidence rate, disease duration at steady state
Relationship among prevalence, incidence rate, disease duration at steady state
Relationship among prevalence, incidence rate, disease duration at steady state
Mean duration of disease
Relationship among prevalence, incidence rate, disease duration at steady state
Relationship among prevalence, incidence rate, disease duration at steady state
Relationship among prevalence, incidence rate, disease duration at steady state
What does steady state mean in the context of estimating P from I and D?
Example varying prevalence, incidence rates and duration of disease
Cross-sectional Bias • Incidence-Prevalence bias – – – – – Type of selection bias If exposed cases have different duration that no-exposed prevalent cases, prevalence ratio will be biased E.g., cases with severe emphysema more likely to smoke, have higher fatality than cases with less severe emphysema, so the prevalence of emphysema in smokers will be underestimated compare to incidence Solution-use incident cases Duration ratio bias – Point prevalence complement ratio bias • Temporal bias – Information bias
Incidence-Prevalence bias • PR 과 – IRR 의 관계 Prev= incidence X duration X (1-prev) PR
* Duration ratio bias * Point prevalence complement ratio bias
Duration ratio bias • • • •
Type of selection bias 드문 질환에서 이환기간이 노출여부와 상관없이 동일하다면 비뚤림 발생하지 않음 노출여부에 따라 질병 이환기간이 다를 때 발생 만성질환의 경우 질병의 duration 이 생존기간과 관련이 있기 때문에 이런 경우 생기는 bias 가 survival bias
Point prevalence complement ratio bias
• • • •
이환기간이 동일하다면 , PR 이 발생 IRR 을 과소측정하는 경향이 노출그룹의 유병률 : 0.04, 비노출그룹 유병률 : 0.01
PR : 4 Point prevalence complement ratio=0.96/0.99=0.97
노출그룹의 유병률 : 0.4, 비노출그룹 유병률 : 0.1
PR : 4 Point prevalence complement ratio=0.6/0.9=0.67
PR, 유병률 크면 → bias 크기 커짐
Selection bias -- Berkson’s bias • • • • • • • Admission-rate bias Cases and/or controls selected from hospitals Result from differential rates of hospital admission for cases and controls If hospital based cases and controls have different exposures that population based, OR will be biased.
E.g., If hospital controls are less likely to have exposures, OR will be over-estimated. E.g., Case control for pancreatic cancer and coffee drinking: Controls were selected from GI patients. However, GI patients are less likely to drink coffee that population. OR was artificially increased. Solution: use population based control, or controls with disease not related to the exposure
Temporal bias • • • • 시간적 선후관계가 모호 – 질병의 위험요인 검정 측면에서의 결정적 단점 – – 예 : 영양결핍과 우울증 연구 시간적 경과에 따른 변동이 없는 노출요인의 경우에는 이러한 제한점에 구애 받지 않음 – 유전적 요인 시간적 선후관계가 뒤집어져 있는 연구는 비추 – 예 : 가설 ) 식이요인이 초경나이에 미치는 영향 대상 ) 중년여성을 대상으로 초경나이와 최근 의 식이습관 조사 전체 유병환자 중 단점을 최소화 Incident cases 만 포함하여 분석함으로 또 다른 bias ?
Historical information 으로 단점 최소화
screening is most likely to pick up less aggressive cancers, because they have a longer interval of being visible on scans while remaining asymptomatic
you find out something earlier but don’t actually change the outcome, and therefore the apparent survival after diagnosis is longer without better survival
Simpson’s paradox aggregated disaggregated
Simpson’s paradox • Aggregated and disaggregated data tell two different stories 치료 종류 환자 수 합계 (n=700) 개복술 경피술 350 350 돌의 크기 < 2cm (n=357) 개복술 87 경피술 270 돌의 크기 ≥ 2cm (n=343) 개복술 263 경피술 80 성 공 273 289 81 234 192 55 실 패 77 61 6 36 71 25 성공률 (%) 78 83 93 87 73 69
단면조사연구 정리 특정 시점 또는 짧은 기간 동안 표본 추출조사 “ 스냅 사진 ” 장점 – 편리하고 비용 효과적 여러 노출과 질병 연구 가능 가설 생성 가능 일반적 인구집단을 대표 단점 시간적 선후관계 모호 생존자만 연구 , 비뚤림 가능 짧은 이환 기간의 질환은 과소측정
Any question?
If you tell the truth you don't have to remember anything. by Mark Twain 1894